ILP clustering - JDK 1.8 about lium

[ahmetaa]

Just FYI, I made a copy of Lium Speaker Diarization code (under GPL) that works with Java-8.

https://github.com/ahmetaa/lium-diarization

The problem was the XStream classes, I updated it and create a build script that can generate a new jar-with-dependencies file. You can open the project under IDEA and run the "stand-alone-jar" target.

[StevenLOL]

Seems that you didn't hit ILP yet,

1. for ILP you can try

sudo apt-get install glpk-utils

to check if you have installed it:

$which glpsol

The error related to ILP is

Exception in thread "main" java.io.IOException: Cannot run program "glpsol": error=2, No such file or directory...

2. did you download UBM and models?

http://www-lium.univ-lemans.fr/diarization/lib/exe/fetch.php/data_ilp.tgz

3. I test all above on UBUNTU 14.04 just now no problem found.

[libeppi]

Thank you for your reply! i checked everything you suggested, and running "tree", gives me the same structure you posted in the homepage. Probably i'm still missing something. I'll let you know.

Thank you.
.
├── go.sh
├── go.sh~
├── ilp_diarization2.sh
├── ilp_diarization2.sh~
├── LICENSE
├── LIUM_SpkDiarization-8.4.1.jar
├── mat
│   ├── wld.efn.xml
│   ├── wld.mahanalobis.mat
│   └── wld.tv.mat
├── models
│   ├── gender.gmms
│   ├── s.gmms
│   ├── sms.gmms
│   └── ubm.gmm
├── README.md
├── t001.out.seg
├── test_out
│   └── t001
│   ├── t001.d.3.seg
│   ├── t001.flt.3.seg
│   ├── t001.g.3.seg
│   ├── t001.gmms
│   ├── t001.h.3.seg
│   ├── t001.init.gmms
│   ├── t001.i.seg
│   ├── t001.l.seg
│   ├── t001.mfcc
│   ├── t001.pms.seg
│   └── t001.s.seg
├── test_wav
│   └── t001.wav
└── ubm
└── wld.gmm

[StevenLOL]

Better check your "wld.efn.xml", download from

http://www-lium.univ-lemans.fr/diarization/lib/exe/fetch.php/data_ilp.tgz ?

[libeppi]

still not working, i just doublechecked and re-downloaded all the files needed. :(

[StevenLOL]

if your └── wld.gmm in folder ubm?

[libeppi]

yes, i re-checked again, nothing yet

[libeppi]

Steven, can you please paste here what happens after:
02:56.304 MatrixIO INFO | read ./mat/wld.tv.mat: row : 3072 col:50 {readRectMatrix() / 1}
02:56.450 TrainIVectorOrTINFO | Compute statistics {make() / 1}
02:56.628 TrainIVectorOrTINFO | Train i-vector {make() / 1}
02:56.629 TotalVariabilitINFO | sent #: 0 {estimateL() / 1}

I think i'm getting closer to the issue. Probably it's not able to open mat/wld.efn.xml
Thank you

[libeppi]

Problem solved - JDK 1.8 was the problem, i downgraded to 1.7 and it is now working

[StevenLOL]

Sorry, the "glpsol1" in ilp_diarization2.sh should be glpsol . I saw your comment on my mobile phone, but I cant't find it on github.

For this issue, maybe it is caused by Java? Or system encoding ? I have this application running on 64bits ubuntu 10.04, 12.04, 12.10 ,14.04, using openjre7 or Oracle java 7, but never encounter this problem. I would like to know your platform, and do you test it on a Visual Machine or Live CD?

[StevenLOL]

Oh,problem solved, thanks libeppi

[libeppi]

👍

[zXpp]

hi,i agree with him that your script is great ,but i met problem when i run the program ,i want to get the output file including: "t001.ilp.problem.120.txt" and "t001.ilp.solition.120.txt" . but when i ./go.sh it wont show these two files .the first is your output files,the second is my tree structure.you can see there are no these two files.I already installed glpk ,and it can work .the last is my result information.so please help slover this problem !thank u

[zXpp]

ok my problem is also solved using jdk 1.7!

[zXpp]

this is the content of the file--t001.ev_is.120.seg.can u tell me what the meaning of the figures and parameters? thank u!

[zXpp]

[libeppi]

Hi zxpp

for example:

9981217_0700_0800_inter_fm_dga 1 1 317 F S U spk0

1 9981217_0700_0800show name
2 1 channel number
3 1 segment start
4 317 lenght of segment
5 F Speaker gender (U unknown, F female, M man)
6 S bandwidth type (T telephone, S studio)
7 U enviroment type (music, speech, ecc)
8 Spk0 speaker label

[zXpp]

thank u ! i got it ,but i still have further questions need your help:
the picutre is another wave file result clusterset;
1.have u noticed that almost every cluster enviroment type were U,but it was music or speech or laugher obviously,why all were detected to unknown?
2 .this is the clusterset reslut,do u know how to run the next step and get the final DER result like the following chart?? the LIUM website didnt mentioned it ,and i am not familiar with java,can u help me ?
thank you !

[StevenLOL]

Hi,@zXpp
If you have the evaluation data, then I think there should have documents on system output format and how to get the DER.

You may find useful tools at NIST :http://www.nist.gov/itl/iad/mig/tools.cfm

[ahmetaa]

So why is it failing in Java 8?